Predictions · Week 3
The model columns show each model's probability for the home team; our pick is The System’s team and its win probability (shown on the featured game and once a game kicks off; sign up for all of them). Games are sorted into the spots our replays found an edge in, with the record behind each. Once a game ends, the models are scored the way players are, and the week and season tallies show who's forecasting best. Imp. is importance: how much the league's playoff picture swings on the game, from simulated seasons (hover for the game's quality).
| Game | Our pick | Elo | Spread | Books | Kalshi | Polymarket | Blend | System | Crowd | Sharp crowd | Imp. | Spot | Result |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ATL at GB | GB 70% ✗ | 66% | 63% | 69% | 69% | 69% | 70% | 70% | — | — | 43 | Tight agreement | ATL 35, GB 14 |
| NYJ at DET | DET 72% ✓ | 69% | 70% | 73% | 74% | 74% | 72% | 72% | 74% | 73% | 37 | Clear favorite, mild disagreement | NYJ 24, DET 31 |
| HOU at IND | HOU 53% ✗ | 48% | 46% | 47% | 47% | 48% | 47% | 47% | 48% | 48% | 47 | Coin flips | HOU 17, IND 19 |
| KC at MIA Bold call of the week | KC 73% ✓ | 48% | 23% | 17% | 16% | 16% | 24% | 27% | 34% | 33% | 41 | Big disagreement | KC 24, MIA 10 |
| TEN at NYG | NYG 57% ✓ | 59% | 57% | 55% | 56% | 56% | 58% | 57% | 61% | 61% | 29 | Tight agreement | TEN 7, NYG 12 |
| CIN at PIT | CIN 58% ✗ | 50% | 40% | 38% | 40% | 40% | 42% | 42% | 45% | 45% | 42 | Big disagreement | CIN 27, PIT 30 |
| SEA at WSH | SEA 76% ✗ | 29% | 26% | 23% | 23% | 23% | 24% | 24% | 42% | 50% | 35 | Clear favorite, mild disagreement | SEA 31, WSH 33 |
| NE at JAX | JAX 57% ✓ | 53% | 59% | 58% | 60% | 60% | 58% | 57% | 68% | 67% | 45 | Everything else | NE 6, JAX 35 |
| LAC at BUF | BUF 78% ✓ | 77% | 70% | 74% | 76% | 76% | 78% | 78% | 79% | 78% | 40 | Tight agreement | LAC 16, BUF 24 |
| CAR at CLE | CAR 52% ✗ | 56% | 43% | 45% | 46% | 47% | 48% | 48% | 46% | 46% | 46 | Big disagreement | CAR 18, CLE 21 |
| ARI at SF Lock of the week | SF 81% ✓ | 78% | 71% | 76% | 78% | 78% | 79% | 81% | 81% | 80% | 42 | Tight agreement | ARI 30, SF 36 |
| MIN at TB | MIN 56% ✓ | 26% | 46% | 49% | 49% | 49% | 43% | 44% | 43% | 43% | 34 | Big disagreement | MIN 23, TB 16 |
| LV at NO | NO 61% ✗ | 52% | 60% | 62% | 65% | 65% | 61% | 61% | 59% | 59% | 54 | Big disagreement | LV 35, NO 27 |
| BAL at DAL | BAL 61% ✓ | 43% | 41% | 39% | 39% | 38% | 39% | 39% | 45% | 44% | 54 | Everything else | BAL 34, DAL 31 |
| LAR at DEN | DEN 51% ✓ | 57% | 54% | 46% | 51% | 51% | 51% | 51% | 48% | 48% | 52 | Everything else | LAR 26, DEN 30 |
| PHI at CHI | PHI 65% ✗ | 39% | 40% | 36% | 37% | 37% | 35% | 35% | 34% | 34% | 46 | Tight agreement | PHI 7, CHI 27 |
| Points this week | +45.0 | +9.6 | -12.0 | -0.7 | +2.3 | +9.4 | +6.0 | +59.1 | +67.2 | on 16 decided games | |||
| Right | 10/16 | 9/16 | 8/16 | 9/16 | 9/16 | 9/16 | 9/16 | 8/15 | 8/15 | ||||
The spots
Each game sorted into the spot the replays found an edge in. Rings are Elo, market and spread; the favorite's logo is in the middle. Tiles turn green or red as games finish.
Tight agreement 5 games · favorite 3 for 5 so far
Lean a little toward the favorite. The blend adds 2 points. Over 27 seasons (2,607 games) favorites here won 69.8% against a 67.6% forecast; in 2023–25 it ran hotter, 74% against 66%.
35–14
GB over ATL ✗
Elo 66% · Mkt 69% · Blend 70%
ATL at GB · Final
7–12
NYG over TEN ✓
Elo 59% · Mkt 56% · Blend 58%
TEN at NYG · Final
16–24
BUF over LAC ✓
Elo 77% · Mkt 75% · Blend 78%
LAC at BUF · Final
30–36
SF over ARI ✓
Elo 78% · Mkt 77% · Blend 79%
ARI at SF · Final
7–27
PHI over CHI ✗
Elo 61% · Mkt 64% · Blend 65%
PHI at CHI · Final
Big disagreement 5 games · favorite 2 for 5 so far
Take the market's number, or split the difference at most. This is where Elo is overconfident. Over 27 seasons (1,102 games) Elo's side won 50% while Elo claimed 65% and the market 48%. The market was right almost every time.
Coin flips 1 game · favorite 0 for 1 so far
Leave the slider near 50. A wrong 65% costs more than a right one earns. Replays: these resolved 50/50, and confident picks on them cost the most points per game.
Clear favorite, mild disagreement 2 games · favorite 1 for 2 so far
The blend without a bump. Nothing to exploit; take the average. Replays: favorites won about as often as the average of the two forecasts said (no surplus either way).
Everything else 3 games · favorite 3 for 3 so far
The blend is as good as anything. No edge found either way in the replays.
Season scoreboard
The models scored as players over every decided game this season, on their pre-kickoff numbers.
| # | Model | Games | Points |
|---|---|---|---|
| 1 | The blend | 119 | +344.5 |
| 2 | Maybe (Kalm) | 119 | +340.5 |
| 3 | Bold (baba) | 119 | +337.5 |
| 4 | The System | 119 | +327.1 |
| 5 | Cautious (baba) | 119 | +321.8 |
| 6 | Elo (the model) | 119 | +307.7 |
| 7 | Pure Elo (baba) | 119 | +307.7 |
| 8 | Presser | 101 | +297.0 |
| 9 | Closing spread | 119 | +295.8 |
| 10 | Pure market (baba) | 119 | +294.7 |
| 11 | Strong (baba) | 119 | +290.1 |
| 12 | QB1 | 101 | +269.0 |
| 13 | The bot crowd | 101 | +267.9 |
| 14 | Scout | 101 | +263.6 |
| 15 | Claudious | 101 | +261.8 |
| 16 | Fader | 101 | +253.0 |
| 17 | Exchange | 101 | +243.9 |
| 18 | Marketeer | 101 | +242.9 |
| 19 | Homer | 101 | +237.2 |
| 20 | The sharp crowd | 46 | +233.3 |
| 21 | The crowd | 46 | +230.4 |
| 22 | Polymarket | 102 | +220.6 |
| 23 | Kalshi | 102 | +218.8 |
| 24 | Sportsbooks | 102 | +211.2 |
| 25 | Contrarian | 101 | +207.7 |